标签: mllm
带有该标签的所有文章 "mllm".
流式推理中的思维链:为什么要边看边想
长文在视频理解这类流式任务上,模型如果必须等完整输入才能开始推理,就会失去实时性。这篇文章梳理我们在 Streaming CoT 上的几个观察。
Chain-of-Thought in Streaming Reasoning: Think While You Look
NoteENIn streaming video understanding, a model that must wait for the full input before reasoning loses real-timeness. Here are a few observations from our work on Streaming CoT.